基于改进的加速鲁棒特征的目标识别

龙思源; 张葆; 宋策; 孙保基

doi:10.3788/CO.20171006.0719

基于改进的加速鲁棒特征的目标识别

doi: 10.3788/CO.20171006.0719

龙思源^{1, 2,},
张葆^1, ,,
宋策¹,
孙保基^{1, 2}

1.
中国科学院长春光学精密机械与物理研究所航测部重点实验室, 吉林长春 130033
2.
中国科学院大学, 北京 100049

基金项目:

中国科学院长春光学精密机械与物理研究所重大创新资助项目 Y3CX1SS14C

详细信息

作者简介:
龙思源(1992-), 男, 湖北襄阳人, 硕士研究生, 主要从事图像处理和目标识别方面的研究。E-mail:Alfred_lsy@163.com

张葆(1966—)，男，吉林磐石人，博士，研究员，博士生导师，主要从事航空光电成像技术方面的研究

通讯作者:
张葆, E-mail:cleresky@vip.sina.com

中图分类号: TP391.4
计量
- 文章访问数: 2477
- HTML全文浏览量: 768
- PDF下载量: 796
- 被引次数: 0
出版历程
- 收稿日期: 2017-05-11
- 修回日期: 2017-07-13
- 刊出日期: 2017-12-01

Object detection based on improved speeded-up robust features

LONG Si-yuan^{1, 2
,},
ZHANG Bao^{1
, ,},
SONG Ce¹,
SUN Bao-ji^{1, 2}

1.
Key Laboratory of Airborne Optical Imaging and Measurement, Changchun Institute of Optics, Fine Mechanics and Physics, Chinese Academy of Sciences, Changchun 130033, China
2.
University of Chinese Academy of Sciences, Beijing 100049, China

Funds:

Major Innovation Project of CIOMP, CAS, China Y3CX1SS14C

More Information

Corresponding author: ZHANG Bao, E-mail:cleresky@vip.sina.com

摘要

摘要: 为了提高加速鲁棒特征（SURF）算法的实时性和准确性，本文提出了一种结合AGAST角点检测和改进的SURF特征描绘算法。首先利用AGAST角点检测模板检测特征点，再使用增加对角信息的哈尔小波响应来生成特征点的描述子，之后利用特征袋对产生的描述子进行编码并生成新的特征向量，最后利用支持向量机（SVM）对特征向量进行分类，完成识别。本文以SIFT和SURF算法为对照，分别进行不同视角、光照和尺度的识别实验。实验结果表明，本文算法的平均识别率为98.0%、96.9%、97.1%，平均时间分别为66.1 ms、79.3 ms、41.0 ms，在识别率上较优于SURF算法，所耗时间约是SURF算法的1/3。
- 图像处理 /
- 目标识别 /
- 加速鲁棒特征 /
- AGAST角点检测
Abstract: To improve the real-time performance and the accuracy of the SURF algorithm, an algorithm combined with AGAST corner detector and improved SURF feature descriptor is proposed. Firstly, feature points are detected by using AGAST corner detection template. Secondly, the Haar wavelet response with increased diagonal information is used to generate descriptor of feature points. Then, the generated descriptor is encoded by the feature bag and a new feature vector is generated. At last, the classification is fulfilled by Support Vector Machine(SVM). Finally, SVM is used to classify the feature vectors to complete the detection. Detection experiments for different view-points, illumination and scales are conducted respectively using SIFT and SURF algorithm as a control. The results show that the average detection rate of this algorithm is 98.0%, 96.9% and 97.1%, and the average time is 66.1 ms, 79.3 ms and 41.0 ms, respectively, which is better than that of SURF algorithm, and the time consumption is about 1/3 of the SURF algorithm.
- image processing /
- object detection /
- Speeded-Up Robust Features /
- AGAST corner detection

HTML全文

图 1 高斯海森矩阵中的L_yy积分模板的近似D_YY

Figure 1. Approximate D_YY of L_yy integral template in Gaussian-Hessian matrix

下载: 全尺寸图片幻灯片

图 2 SURF描述子的生成示意图

Figure 2. Schematic diagram of SURF descriptors generating

下载: 全尺寸图片幻灯片

图 3 AGAST在不同尺度下的层和内层

Figure 3. Octaves and intra-octaves of AGAST algorithm in different scales

下载: 全尺寸图片幻灯片

图 4 Haar特征模板

Figure 4. Scores of Haar feature

下载: 全尺寸图片幻灯片

图 5 识别算法流程

Figure 5. Flowchart of proposed detection algorithm

下载: 全尺寸图片幻灯片

图 6 不同角度的样本目标

Figure 6. Objects of different vew-points

下载: 全尺寸图片幻灯片

图 7 不同光照的样本目标

Figure 7. Objects of different illumination

下载: 全尺寸图片幻灯片

表 1 不同视角下3种算法的性能比较

Table 1. Performances comparison of three algorithms under different view-points

	SIFT	SURF	本文算法
正确数	102	99	100
正确率/%	100	97.0	98.0
平均时间/ms	568.7	287.2	66.1

下载: 导出CSV

表 2 不同光照下3种算法的性能比较

Table 2. Performances comparison of three algorithms under different illumination

	SIFT	SURF	本文算法
正确数	38	47	47
正确率/%	77.6	95.9	95.9
平均时间/ms	590.6	272.9	79.3

下载: 导出CSV

表 3 不同尺度下3种算法的性能比较

Table 3. Performances comparison of three algorithms under different scales

	SIFT	SURF	本文算法
正确数	98	96	98
正确率/%	96.1	94.1	96.1
平均时间/ms	294.1	146.2	41.0

下载: 导出CSV

参考文献(15)

[1]	贾平, 徐宁, 张叶.基于局部特征提取的目标自动识别[J].光学精密工程, 2013, 21(7):1898-1905. http://kns.cnki.net/KCMS/detail/detail.aspx?filename=gxjm201307035&dbname=CJFD&dbcode=CJFQ JIA P, XU N, ZHANG Y. Automatic target recognition based on local feature extraction[J]. Opt. Precision Eng., 2013, 21(7):1898-1905.(in Chinese) http://kns.cnki.net/KCMS/detail/detail.aspx?filename=gxjm201307035&dbname=CJFD&dbcode=CJFQ
[2]	赵春阳, 赵怀慈.多模态鲁棒的局部特征描述符[J].光学精密工程, 2015, 23(5):1474-1483. http://kns.cnki.net/KCMS/detail/detail.aspx?filename=gxjm201505035&dbname=CJFD&dbcode=CJFQ ZHAO CH Y, ZHAO H C. Multimodality robust local feature descriptors[J]. Opt. Precision Eng., 2015, 23(5):1474-1483.(in Chinse) http://kns.cnki.net/KCMS/detail/detail.aspx?filename=gxjm201505035&dbname=CJFD&dbcode=CJFQ
[3]	LOWE D. Distinctive image features from scale-invariant keypoints[J]. International J. Computer Vision, 2004, 60(2):91-110. doi: 10.1023/B:VISI.0000029664.99615.94
[4]	HARRIS C, STEPHENS M. A combined corner and edge detector[J]. Alvey Vision Conference, 1988(3):147-151. http://ci.nii.ac.jp/naid/10007785664/
[5]	BAY H, TUYTELAARS T, VAN GOOL L. SURF:speeded-up robust features[J]. Computer Vision & Image Understanding, 2006, 110(3):404-417. http://www.springerlink.com/index/e580h2k58434p02k.pdf
[6]	MIKOLAJCZYK K, SCHMID C. A performance evaluation of local descriptors[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2005, 27(10):1615-1630. doi: 10.1109/TPAMI.2005.188
[7]	王飞宇, 邸男, 贾平.结合尺度空间FAST角点检测器和SURF描绘器的图像特征[J].液晶与显示, 2014, 29(4):598-604. http://kns.cnki.net/KCMS/detail/detail.aspx?filename=yjys201404025&dbname=CJFD&dbcode=CJFQ WANG F Y, DI N, JIA P. Image features using scale-space FAST corner detector and SURF descriptor[J]. Chinese J. Liquid Crystals and Displays, 2014, 29(4):598-604.(in Chinese) http://kns.cnki.net/KCMS/detail/detail.aspx?filename=yjys201404025&dbname=CJFD&dbcode=CJFQ
[8]	成培瑞, 王建立, 王斌, 等.基于多尺度区域对比的显著目标识别[J].中国光学, 2016, 9(1):97-105. http://www.chineseoptics.net.cn/CN/abstract/abstract9392.shtml CHENG P R, WANG J L, WANG B, et al.. Salient object detection based on multi-scale region contrast[J]. Chinese Optics, 2016, 9(1):97-105.(in Chinese) http://www.chineseoptics.net.cn/CN/abstract/abstract9392.shtml
[9]	武治国, 李桂菊.动态目标识别中的实时复杂巡航场景运动检测[J].液晶与显示, 2014, 29(5):844-849. http://kns.cnki.net/KCMS/detail/detail.aspx?filename=yjys201405036&dbname=CJFD&dbcode=CJFQ WU ZH G, LI G J. Real-time complex cruise scene detection technology in target recognition[J]. Chinese J. Liquid Crystals and Displays, 2014, 29(5):844-849.(in Chinese) http://kns.cnki.net/KCMS/detail/detail.aspx?filename=yjys201405036&dbname=CJFD&dbcode=CJFQ
[10]	MAIR E, HAGER G, BURSCHKA D, et al.. Adaptive and generic corner detection based on the accelerated segment test[J]. Computer Vision-ECCV, 2010, 2010:183-196. doi: 10.1007/978-3-642-15552-9_14
[11]	王丽.融合底层和中层字典特征的行人重识别[J].中国光学, 2016, 9(5):540-546. http://www.chineseoptics.net.cn/CN/abstract/abstract9441.shtml WANG L. Pedestrian re-identification based on fusing low-level and mid-level features[J]. Chinese Optics, 2016, 9(5):540-546.(in Chinese) http://www.chineseoptics.net.cn/CN/abstract/abstract9441.shtml
[12]	ZHANG H, HU Q. Fast image matching based-on improved SURF algorithm[C]. International Conference on Electronics, Communications and Automation Control, IEEE, 2011:1460-1463.
[13]	ROSTEN E, PORTER R B, DRUMMOND T, et al.. Faster and better:a machine learning approach to corner detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2010, 32(1):105-119. doi: 10.1109/TPAMI.2008.275
[14]	CSURKA G, DANCE C, FAN L, et al.. Visual categorization with bags of keypoints[C]. 8th European Conference on Computer Vision, Progue, Czech Republic, 2004:1-22.
[15]	DONGHOON K, ROZENN D. Face components detection using SURF descriptors and SVMs[C]. IEEE International Machine Vision and Image Processing Conference, IEEE, 2008:51-56.